A stochastic backpropagation algorithm for training neural networks

被引：9

作者：

Chen, YQ ^{[1
]}

Yin, T ^{[1
]}

Babri, HA ^{[1
]}

机构：

[1] Nanyang Technol Univ, Sch Elect & Elect Engn, Singapore 639798, Singapore

来源：

ICICS - PROCEEDINGS OF 1997 INTERNATIONAL CONFERENCE ON INFORMATION, COMMUNICATIONS AND SIGNAL PROCESSING, VOLS 1-3: THEME: TRENDS IN INFORMATION SYSTEMS ENGINEERING AND WIRELESS MULTIMEDIA COMMUNICATIONS | 1997年

关键词：

D O I：

10.1109/ICICS.1997.652068

中图分类号：

TP [自动化技术、计算机技术];

学科分类号：

0812 ;

摘要：

The popularly used backpropagation algorithm (BP) for training multilayered neural networks is generally slow and prone to getting stuck in local minima. In this paper, a novel method to improve the performance of BP by randomising the cost function is proposed. The method is effective in helping the BP algorithm to escape from local minima and therefore improve convergence and generalization. This is demonstrated on a non-convex pattern recognition problem.

引用

页码：703 / 707

页数：3